[cpp-threads] memory model

[Peter Dimov]

Hans Boehm wrote:
> The intent is yes in all cases.  It is currently an open problem how
> to state that in the current memory model.  But I don't think anyone
> has thought about it for very long, yet.

But the current wording doesn't allow reordering in (2) and (3), if I'm not 
mistaken?

> Hans
>
> On Thu, 5 May 2005, Peter Dimov wrote:
>
>> I have the following question about the proposed memory model:
>>
>> In the examples below:
>>
>> 1.
>>
>>     r1 = atomic_load( &x );
>>     r2 = y;
>>
>> 2.
>>
>>     r1 = atomic_load( &x );
>>     r2 = atomic_load( &y );
>>
>> 3.
>>
>>     if( r1 = atomic_load( &x ) )
>>         r2 = y;
>>
>> (note that the atomic loads do not have an associated ordering
>> constraint)
>>
>> is the compiler or the hardware allowed to reorder the two loads?